All,
just to share two noteworthy things:
- There is now a new tutorial program, step-35, that solves the Navier-Stokes
equations using a completely decoupled (projection) scheme. The program was
contributed by Abner Salgado-Gonzalez -- thanks! -- and the program is here:
http://dealii.org/developer/doxygen/deal.II/step_35.html
- Timo Heister (mostly) and I (a little bit) have been busy working in the
last couple of weeks on a scheme to distribute meshes across the
nodes of a cluster communicating over MPI. As you know, previously
we could already compute using multiple nodes, but the mesh and the
DoFHandler had to be held on each processor, providing the bottleneck
that limited how far we could go in parallel.
With the new scheme, which builds on the p4est library written by
Carsten Burstedde and Lucas Wilcox at UTexas at Austin, we can
make sure that each processor only holds slightly more cells and degrees
of freedom than it actually needs. The biggest runs we've done so far had
over 1 billion (!) cells in both 2d and 3d, using 120 processors. That's on
the order of 50 times bigger meshes than we've ever generated.
Much remains to be done before we can actually do computations on such
meshes, but it's a good start and we've got plenty of ideas of where to go
from here (and problems: on a billion cells, you'll quickly have more
degrees of freedom than there are numbers in an unsigned int ;-)
This much for now, stay tuned for more news soon!
Best
Wolfgang
--
-------------------------------------------------------------------------
Wolfgang Bangerth email: [email protected]
www: http://www.math.tamu.edu/~bangerth/
_______________________________________________
dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
